Is your graphics LCD one of the COG (chip on glass) types that has the fine pitch flex cable coming down to the PC board? If so has the host PC board got a series of some 7 or 8 capacitors to support the LCD voltage supply to the COG display?

If you say yes I can almost guess what your problem is. I have seen an assorted batch of problems with GLCD modules of the COG type wherein the capacitors specified for use in the controller data sheet are way too big. It is common for the controller data sheets to recommend capacitors in the .47uF to 4.7uF size range. I have found that it is necessary to scale these values back to .1uF to 1.0uF in order to get reliable startup and re-init of the controller to work correctly. The charge pump circuits on the GLCD controller chips can create excessive "GND" noise up on the COG chip mount area due to the caps being 4 to 6 cm away from the glass on the other end of a 0.5 mm pitch flex cable. This noise can be so extreme that the microcontroller trying to initialize the display chip will have trouble initializing the chip with resultant register corruption. Smaller caps solve this problem by lowering the amount of noise generated.

COG displays are awsome in terms of slim lines and sleek design but they are far from an optimum design when the caps are too far away from the controller chip.

-----------------------

Another issue that I have found is that sometimes it is necessary to invoke the /RESET signal pulse to the COG display for an extended period of time before the LCD controller initialize. This is necessary for the vary reason you are having problems. At normal first power on the LCD controller has its own built in POR that makes a 'too short reset' from the microcontroller init transparent if it is not long enough. Then at the time you reset the microcontroller via a switch the LCD controller does not see a POR and the "too short reset" generated by the software initialize becomes too short.

In one SiLabs C8051F126 design I worked on with a GLCD/COG module I arranged the /RESET signal to the LCD cable to be pulled down via a resistor to GND. This resistor (about 1K) would hold the RESET active from the time the microcontroller went into reset until such time that the software got around to making the pin be an output and pulling it up (I was using the SiLabs part in its I/O pin being in the push pull mode). This lets the LCD reset period be the length of the microcontroller reset pulse plus the start up delay of microcontroller, which included a FLASH checksum verify, plus whatever additional reset time the microcontroller software allowed for in the LCD init routine.

Michael Karas